White
WhiteMat = whiteningMatrix(cov(noclassData), method="PCA")
sum(is.na(WhiteMat))
[1] 0
tokeep <- apply(is.na(WhiteMat),1,sum) == 0
WhiteMat <- WhiteMat[tokeep,]
sum(is.na(WhiteMat))
[1] 0
sum(apply(abs(WhiteMat),1,sum) > 1.0e6)
[1] 0
tokeep <- apply(abs(WhiteMat),1,sum) < 1.0e6
WhiteMat <- WhiteMat[tokeep,]
sum(apply(abs(WhiteMat),1,sum) > 1.0e6)
[1] 0
pander::pander(c(ncol=ncol(WhiteMat),nrow=nrow(WhiteMat)))
pander::pander(WhiteMat[1:TopVariables,1:TopVariables])
| L1 |
10.13 |
-1.36 |
-3.37 |
-0.210 |
-0.17 |
| L2 |
10.40 |
4.13 |
15.96 |
1.175 |
1.05 |
| L3 |
-9.24 |
-26.95 |
12.28 |
-0.274 |
1.24 |
| L4 |
3.87 |
-26.10 |
-61.25 |
6.672 |
6.08 |
| L5 |
-4.87 |
27.71 |
-45.41 |
3.208 |
1.69 |
PCAWhite_Train <- as.data.frame(cbind(tcrossprod(as.matrix(noclassData), WhiteMat),trainDataFrame[,notcorr]))
colnames(PCAWhite_Train) <- c(colnames(tcrossprod(as.matrix(noclassData), WhiteMat)),notcorr)
sum(is.na(PCAWhite_Train))
[1] 0
PCAWhitePredicted <- as.data.frame(cbind(tcrossprod(as.matrix(noclassTestData), WhiteMat),testDataFrame[,notcorr]))
colnames(PCAWhitePredicted) <- c(colnames(tcrossprod(as.matrix(noclassTestData), WhiteMat)),notcorr)
sum(is.na(PCAWhitePredicted))
[1] 0
iscontinous <- colnames(PCAWhitePredicted)[sapply(apply(PCAWhitePredicted,2,unique),length) > 5] ## Only variables with enough
varlistWhite <- iscontinous
univarWhite <- uniRankVar(varlistWhite,
paste(outcome,"~1"),
outcome,
PCAWhite_Train,
rankingTest = "CStat")
100 : M_ST129TS 200 : D_ST50TA 300 : D_ST48CV
univarWhiteTest <- uniRankVar(varlistWhite,
paste(outcome,"~1"),
outcome,
PCAWhite_Train,
testData=PCAWhitePredicted,
rankingTest = "CStat")
100 : M_ST129TS 200 : D_ST50TA 300 : D_ST48CV
univarWhite$orderframe$BACC <- (univarWhite$orderframe$Sensitivity + univarWhite$orderframe$Specificity)/2.0
univarWhiteTest$orderframe$BACC <- (univarWhiteTest$orderframe$Sensitivity + univarWhiteTest$orderframe$Specificity)/2.0
Correlation
Matrices
RAW
par(cex=1.0,cex.main=0.8)
breaks <- c(0:5)/5.0;
cormat <- cor(testDataFrame,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(max(abs(cormat)))
1
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.294 |
0.433 |
0.535 |
0.627 |
0.792 |
pander::pander(c(Raw_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Raw Correlation",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature"
)

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Raw Correlation",xlab="Spearman Correlation")
rawDen <- density(cormat,from=-1,to=1)
par(op)
UPSTM Blind
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(DEdataframe,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
#pander::pander(colnames(cormat)[apply(abs(cormat),2,max)>rhoThreshold])
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.144 |
0.2 |
0.265 |
0.353 |
0.536 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(predTestDe,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.145 |
0.202 |
0.269 |
0.359 |
0.538 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after HCA",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after UPSTM",xlab="Spearman Correlation")
DeDen <- density(cormat,from=-1,to=1)
par(op)
UPSTM
Blind/Spearman
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(DEdataframeSpear,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
#pander::pander(colnames(cormat)[apply(abs(cormat),2,max)>rhoThreshold])
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.143 |
0.195 |
0.257 |
0.347 |
0.542 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(predTestDeSpear,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.142 |
0.197 |
0.264 |
0.353 |
0.54 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after HCA:Spearman",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after UPSTM",xlab="Spearman Correlation")
DeSpearDen <- density(cormat,from=-1,to=1)
par(op)
UPSTM Driven
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(DriDEdataframe,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
#pander::pander(colnames(cormat)[apply(abs(cormat),2,max)>rhoThreshold])
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.15 |
0.209 |
0.276 |
0.369 |
0.537 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(DriDEdataframe,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.147 |
0.205 |
0.271 |
0.36 |
0.527 |
pander::pander(c(HCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after Driven-HCA",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after Driven-UPSTM",xlab="Spearman Correlation")
DeDrivDen <- density(cormat,from=-1,to=1)
par(op)
UPSTM
Spearman
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(DriDEdataframeSpear,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
#pander::pander(colnames(cormat)[apply(abs(cormat),2,max)>rhoThreshold])
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.155 |
0.216 |
0.287 |
0.379 |
0.553 |
pander::pander(c(HCAS_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(predTestDriSpear,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.157 |
0.225 |
0.296 |
0.387 |
0.556 |
pander::pander(c(HCAS_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation: Driven/Spearman",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after UPSTM with Spearman",xlab="Spearman Correlation")
DeDrivSpearDen <- density(cormat,from=-1,to=1)
par(op)
PCA
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(PCA_Train,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.152 |
0.22 |
0.309 |
0.42 |
0.707 |
pander::pander(c(PCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(PCA_Predicted,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.148 |
0.223 |
0.312 |
0.428 |
0.688 |
pander::pander(c(PCA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after PCA",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after PCA",xlab="Spearman Correlation")
PCADen <- density(cormat,from=-1,to=1)
par(op)
EFA
par(cex=1.0,cex.main=0.8)
## Train Correlation
cormat <- cor(EFA_Train,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.152 |
0.22 |
0.309 |
0.42 |
0.707 |
pander::pander(c(EFA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(EFA_Predicted,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.148 |
0.223 |
0.311 |
0.428 |
0.687 |
pander::pander(c(EFA_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after EFA",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after EFA",xlab="Spearman Correlation")
EFADen <- density(cormat,from=-1,to=1)
par(op)
PCA Whitening
## Train Correlation
cormat <- cor(PCAWhite_Train,method="pearson")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Train=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.136 |
0.192 |
0.273 |
0.383 |
0.653 |
pander::pander(c(PCAWhite_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
## Test Correlation
cormat <- cor(PCAWhitePredicted,method="spearman")
cormat[is.na(cormat)] <- 0
diag(cormat) <- 0;
pander::pander(c(Test=max(abs(cormat))))
pander::pander(quantile(abs(cormat),probs=c(0.9,0.95,0.975,0.99,0.999)))
| 0.138 |
0.196 |
0.273 |
0.387 |
0.633 |
pander::pander(c(PCAWhite_fraction=sum(abs(cormat)>rhoThreshold)/ncol(cormat)/ncol(cormat)))
gplots::heatmap.2(abs(cormat),
trace = "none",
mar = c(5,5),
col=rev(heat.colors(5)),
breaks = breaks,
main = "Correlation after PCAWhite",
cexRow = 0.25,
cexCol = 0.25,
srtCol=35,
srtRow=75,
key.title=NA,
key.xlab="Spearman Correlation",
xlab="Feature", ylab="Feature")

#hist(cormat,freq=FALSE,
# density=NULL,
# xlim=c(-1,1),
# ylim=c(0,4.0),
# main="Correlation after PCAWhite",xlab="Spearman Correlation")
WhiteDen <- density(cormat,from=-1,to=1)
par(op)
The Density
Plot
par(cex=0.7)
colors=c("red","blue","green","darkblue","darkgreen","purple","orange","darkred");
plot(rawDen,
xlim=c(-1,1),
ylim=c(0,7.0),
col=colors[1],
lty=1,
lwd=4,
main="Test: Correlation Distribution",xlab="Spearman Correlation")
lines(DeDen,col=colors[2],lty=2,lwd=4)
lines(DeSpearDen,col=colors[3],lty=3,lwd=4)
lines(DeDrivDen,col=colors[4],lty=4,lwd=2)
lines(DeDrivSpearDen,col=colors[5],lty=5,lwd=2)
lines(PCADen,col=colors[6],lty=6,lwd=1)
lines(EFADen,col=colors[7],lty=7,lwd=1)
lines(WhiteDen,col=colors[8],lty=8,lwd=1)
names=c("Raw","HCA:P","HCA:S","DHCA:P","DHCA:S","PCA","EFA","White:PCA")
#colors=c("red","blue","green","blue","green","purple","purple","gray");
lines=c(1,2,3,4,5,6,7,8)
lwds=c(4,4,4,2,2,1,1,1)
legend("topleft",names,col=colors,lty=lines,lwd=lwds,cex=0.50)

par(op)
Differences
between train and test ROC AUC
par(op)
par(mfrow=c(1,1),cex=0.7)
AUCResults <- list();
diffAUC <- list();
thenames <- rownames(univar$orderframe)[(rownames(univar$orderframe) %in% colnames(testDataFrame))]
rawAUC <- univar$orderframe[thenames,"ROCAUC"]
thenames <- thenames[rawAUC >= aucTHR]
rawAUC <- univar$orderframe[thenames,"ROCAUC"]
rawAUCTest <- univarTest$orderframe[thenames,"ROCAUC"]
AUCResults$RAW <- rawAUCTest
diffAUC$RAW <- rawAUCTest-rawAUC
plot(rawAUC,rawAUCTest-rawAUC,
xlab="TRAIN ROC AUC",
ylab="Test:AUC-Train:AUC",
xlim=c(0.5,1.0),
ylim=c(-0.25,0.25),
pch=1,
col=colors[1],
main="ROC AUC Difference Between Test and Train")
thenames <- rownames(univarDe$orderframe)[!(rownames(univarDe$orderframe) %in% colnames(testDataFrame))]
HCAP <- univarDe$orderframe[thenames,"ROCAUC"]
thenames <- thenames[HCAP >= aucTHR]
HCAP <- univarDe$orderframe[thenames,"ROCAUC"]
HCAPTest <- univarDeTest$orderframe[thenames,"ROCAUC"]
AUCResults$HCAP <- HCAP
AUCResults$HCAP_T <- HCAPTest
diffAUC$HCAP <- HCAPTest-HCAP
points(HCAP,HCAPTest-HCAP,pch=2,col=colors[2])
thenames <- rownames(univarDeSpear$orderframe)[!(rownames(univarDeSpear$orderframe) %in% colnames(testDataFrame))]
HCAS <- univarDeSpear$orderframe[thenames,"ROCAUC"]
thenames <- thenames[HCAS >= aucTHR]
HCAS <- univarDeSpear$orderframe[thenames,"ROCAUC"]
HCASTest <- univarDeSpearTest$orderframe[thenames,"ROCAUC"]
AUCResults$HCAS <- HCAS
AUCResults$HCAS_T <- HCASTest
diffAUC$HCAS <- HCASTest-HCAS
points(HCAS,HCASTest-HCAS,pch=3,col=colors[3])
thenames <- rownames(univarDeDri$orderframe)[!(rownames(univarDeDri$orderframe) %in% colnames(testDataFrame))]
DHCAP <- univarDeDri$orderframe[thenames,"ROCAUC"]
thenames <- thenames[DHCAP >= aucTHR]
DHCAP <- univarDeDri$orderframe[thenames,"ROCAUC"]
DHCAPTest <- univarDeDriTest$orderframe[thenames,"ROCAUC"]
AUCResults$DHCAP <- DHCAP
AUCResults$DHCAP_T <- DHCAPTest
diffAUC$DHCAP <- DHCAPTest-DHCAP
points(DHCAP,DHCAPTest-DHCAP,pch=4,col=colors[4])
thenames <- rownames(univarDeDriSpear$orderframe)[!(rownames(univarDeDriSpear$orderframe) %in% colnames(testDataFrame))]
DHCAS <- univarDeDriSpear$orderframe[thenames,"ROCAUC"]
thenames <- thenames[DHCAS >= aucTHR]
DHCAS <- univarDeDriSpear$orderframe[thenames,"ROCAUC"]
DHCASTest <- univarDeDriSpearTest$orderframe[thenames,"ROCAUC"]
AUCResults$DHCAS <- DHCAS
AUCResults$DHCAS_T <- DHCASTest
diffAUC$DHCAS <- DHCASTest-DHCAS
points(DHCAS,DHCASTest-DHCAS,pch=5,col=colors[5])
thenames <- rownames(univarPCA$orderframe)[!(rownames(univarPCA$orderframe) %in% colnames(testDataFrame))]
PCA <- univarPCA$orderframe[thenames,"ROCAUC"]
thenames <- thenames[PCA >= aucTHR]
PCA <- univarPCA$orderframe[thenames,"ROCAUC"]
PCATest <- univarPCATest$orderframe[thenames,"ROCAUC"]
AUCResults$PCA <- PCA
AUCResults$PCA_T <- PCATest
diffAUC$PCA <- PCATest-PCA
points(PCA,PCATest-PCA,pch=6,col=colors[6])
thenames <- rownames(univarEFA$orderframe)[!(rownames(univarEFA$orderframe) %in% colnames(testDataFrame))]
EFA <- univarEFA$orderframe[thenames,"ROCAUC"]
thenames <- thenames[EFA >= aucTHR]
EFA <- univarEFA$orderframe[thenames,"ROCAUC"]
EFATest <- univarEFATest$orderframe[thenames,"ROCAUC"]
AUCResults$EFA <- EFA
AUCResults$EFA_T <- EFATest
diffAUC$EFA <- EFATest-EFA
points(EFA,EFATest-EFA,pch=7,col=colors[7])
thenames <- rownames(univarWhite$orderframe)[!(rownames(univarWhite$orderframe) %in% colnames(testDataFrame))]
WPCA <- univarWhite$orderframe[thenames,"ROCAUC"]
thenames <- thenames[WPCA >= aucTHR]
WPCA <- univarWhite$orderframe[thenames,"ROCAUC"]
WPCATest <- univarWhiteTest$orderframe[thenames,"ROCAUC"]
AUCResults$WPCA <- WPCA
AUCResults$WPCA_T <- WPCATest
diffAUC$WPCA <- WPCATest-WPCA
points(WPCA,WPCATest-WPCA,pch=8,col=colors[8])
names=c("Raw","HCA:P","HCA:S","DHCA:P","DHCA:S","PCA","EFA","White:PCA")
pchs=c(1,2,3,4,5,6,7,8)
legend("bottomright",names,col=colors,pch=pchs,cex=0.50)

par(op)
Violin of
differences
par(op)
par(mfrow=c(1,1),cex=0.7)
vioplot(diffAUC,
ylim=c(-0.25,0.25),
ylab="Test-Train",
main="Test-Train Paired ROC AUC Difference",
col=colors,
cex.axis=0.6,
las=2
)
stripchart(diffAUC, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(diffAUC),lapply(diffAUC,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
Distribution of
ROC AUC in latent Variables
par(op)
par(mfrow=c(1,1),cex=0.7)
colors2 <- length(AUCResults)
colors2[1] <- colors[1];
colors2[2*(1:(length(colors)-1))] <- colors[2:length(colors)]
colors2[1+2*(1:(length(colors)-1))] <- colors[2:length(colors)]
vioplot(AUCResults,
ylim=c(0.3,1.0),
ylab="ROC AUC",
main="ROC AUC of Latent Variables",
col=colors2,
cex.axis=0.6,
las=2
)
abline(h=0.5,col="black")
stripchart(AUCResults, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(AUCResults),lapply(AUCResults,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
Differences
between train and test Balanced Accuracy
par(op)
par(mfrow=c(1,1),cex=0.7)
BACCTHR <- aucTHR
BACCResults <- list();
diffBACC <- list();
thenames <- rownames(univar$orderframe)[(rownames(univar$orderframe) %in% colnames(testDataFrame))]
rawBACC <- univar$orderframe[thenames,"BACC"]
thenames <- thenames[rawBACC >= BACCTHR]
rawBACC <- univar$orderframe[thenames,"BACC"]
rawBACCTest <- univarTest$orderframe[thenames,"BACC"]
BACCResults$RAW <- rawBACCTest
diffBACC$RAW <- rawBACCTest-rawBACC
plot(rawBACC,rawBACCTest-rawBACC,
xlab="TRAIN Balanced Acc",
ylab="Test:BACC-Train:BACC",
xlim=c(0.5,1.0),
ylim=c(-0.25,0.25),
pch=1,
col=colors[1],
main="Balanced Acc Difference Between Test and Train")
thenames <- rownames(univarDe$orderframe)[!(rownames(univarDe$orderframe) %in% colnames(testDataFrame))]
HCAP <- univarDe$orderframe[thenames,"BACC"]
thenames <- thenames[HCAP >= BACCTHR]
HCAP <- univarDe$orderframe[thenames,"BACC"]
HCAPTest <- univarDeTest$orderframe[thenames,"BACC"]
BACCResults$HCAP <- HCAP
BACCResults$HCAP_T <- HCAPTest
diffBACC$HCAP <- HCAPTest-HCAP
points(HCAP,HCAPTest-HCAP,pch=2,col=colors[2])
thenames <- rownames(univarDeSpear$orderframe)[!(rownames(univarDeSpear$orderframe) %in% colnames(testDataFrame))]
HCAS <- univarDeSpear$orderframe[thenames,"BACC"]
thenames <- thenames[HCAS >= BACCTHR]
HCAS <- univarDeSpear$orderframe[thenames,"BACC"]
HCASTest <- univarDeSpearTest$orderframe[thenames,"BACC"]
BACCResults$HCAS <- HCAS
BACCResults$HCAS_T <- HCASTest
diffBACC$HCAS <- HCASTest-HCAS
points(HCAS,HCASTest-HCAS,pch=3,col=colors[3])
thenames <- rownames(univarDeDri$orderframe)[!(rownames(univarDeDri$orderframe) %in% colnames(testDataFrame))]
DHCAP <- univarDeDri$orderframe[thenames,"BACC"]
thenames <- thenames[DHCAP >= BACCTHR]
DHCAP <- univarDeDri$orderframe[thenames,"BACC"]
DHCAPTest <- univarDeDriTest$orderframe[thenames,"BACC"]
BACCResults$DHCAP <- DHCAP
BACCResults$DHCAP_T <- DHCAPTest
diffBACC$DHCAP <- DHCAPTest-DHCAP
points(DHCAP,DHCAPTest-DHCAP,pch=4,col=colors[4])
thenames <- rownames(univarDeDriSpear$orderframe)[!(rownames(univarDeDriSpear$orderframe) %in% colnames(testDataFrame))]
DHCAS <- univarDeDriSpear$orderframe[thenames,"BACC"]
thenames <- thenames[DHCAS >= BACCTHR]
DHCAS <- univarDeDriSpear$orderframe[thenames,"BACC"]
DHCASTest <- univarDeDriSpearTest$orderframe[thenames,"BACC"]
BACCResults$DHCAS <- DHCAS
BACCResults$DHCAS_T <- DHCASTest
diffBACC$DHCAS <- DHCASTest-DHCAS
points(DHCAS,DHCASTest-DHCAS,pch=5,col=colors[5])
thenames <- rownames(univarPCA$orderframe)[!(rownames(univarPCA$orderframe) %in% colnames(testDataFrame))]
PCA <- univarPCA$orderframe[thenames,"BACC"]
thenames <- thenames[PCA >= BACCTHR]
PCA <- univarPCA$orderframe[thenames,"BACC"]
PCATest <- univarPCATest$orderframe[thenames,"BACC"]
BACCResults$PCA <- PCA
BACCResults$PCA_T <- PCATest
diffBACC$PCA <- PCATest-PCA
points(PCA,PCATest-PCA,pch=6,col=colors[6])
thenames <- rownames(univarEFA$orderframe)[!(rownames(univarEFA$orderframe) %in% colnames(testDataFrame))]
EFA <- univarEFA$orderframe[thenames,"BACC"]
thenames <- thenames[EFA >= BACCTHR]
EFA <- univarEFA$orderframe[thenames,"BACC"]
EFATest <- univarEFATest$orderframe[thenames,"BACC"]
BACCResults$EFA <- EFA
BACCResults$EFA_T <- EFATest
diffBACC$EFA <- EFATest-EFA
points(EFA,EFATest-EFA,pch=7,col=colors[7])
thenames <- rownames(univarWhite$orderframe)[!(rownames(univarWhite$orderframe) %in% colnames(testDataFrame))]
WPCA <- univarWhite$orderframe[thenames,"BACC"]
thenames <- thenames[WPCA >= BACCTHR]
WPCA <- univarWhite$orderframe[thenames,"BACC"]
WPCATest <- univarWhiteTest$orderframe[thenames,"BACC"]
BACCResults$WPCA <- WPCA
BACCResults$WPCA_T <- WPCATest
diffBACC$WPCA <- WPCATest-WPCA
points(WPCA,WPCATest-WPCA,pch=8,col=colors[8])
names=c("Raw","HCA:P","HCA:S","DHCA:P","DHCA:S","PCA","EFA","White:PCA")
pchs=c(1,2,3,4,5,6,7,8)
legend("bottomright",names,col=colors,pch=pchs,cex=0.50)

par(op)
Violin of
differences
par(op)
par(mfrow=c(1,1),cex=0.7)
vioplot(diffBACC,
ylim=c(-0.25,0.25),
ylab="Test-Train",
main="Test-Train Paired Balanced Acc Difference",
col=colors,
cex.axis=0.6,
las=2
)
stripchart(diffBACC, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(diffBACC),lapply(diffBACC,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
Distribution of
Balanced Acc in latent Variables
par(op)
par(mfrow=c(1,1),cex=0.7)
colors2 <- length(BACCResults)
colors2[1] <- colors[1];
colors2[2*(1:(length(colors)-1))] <- colors[2:length(colors)]
colors2[1+2*(1:(length(colors)-1))] <- colors[2:length(colors)]
vioplot(BACCResults,
ylim=c(0.3,1.0),
ylab="Balanced Acc",
main="Balanced Acc of Latent Variables",
col=colors2,
cex.axis=0.6,
las=2
)
abline(h=0.5,col="black")
stripchart(BACCResults, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(BACCResults),lapply(BACCResults,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
Differences
between train and test IDI
par(op)
par(mfrow=c(1,1),cex=0.7)
testIDI <- list();
diffIDI <- list();
rawIDI <- univar$orderframe$IDI
rawIDITest <- univarTest$orderframe$IDI
testIDI$RAW <- rawIDITest
diffIDI$RAW <- rawIDITest-rawIDI
plot(rawIDI,rawIDITest-rawIDI,
xlab="TRAIN Test IDI",
ylab="Test:IDI-Train:IDI",
xlim=c(0,0.5),
ylim=c(-0.2,0.2),
pch=1,
col=colors[1],
main="Predict IDI Difference Between Test and Train")
HCAP <- univarDe$orderframe$IDI[!(rownames(univarDe$orderframe) %in% colnames(testDataFrame))]
HCAPTest <-univarDeTest$orderframe$IDI[!(rownames(univarDe$orderframe) %in% colnames(testDataFrame))]
testIDI$HCAP <- HCAP
testIDI$HCAP_T <- HCAPTest
diffIDI$HCAP <- HCAPTest-HCAP
points(HCAP,HCAPTest-HCAP,pch=2,col=colors[2])
HCAS <- univarDeSpear$orderframe$IDI[!(rownames(univarDeSpearTest$orderframe) %in% colnames(testDataFrame))]
HCASTest <- univarDeSpearTest$orderframe$IDI[!(rownames(univarDeSpearTest$orderframe) %in% colnames(testDataFrame))]
testIDI$HCAS <- HCAS
testIDI$HCAS_T <- HCASTest
diffIDI$HCAS <- HCASTest-HCAS
points(HCAS,HCASTest-HCAS,pch=3,col=colors[3])
DHCAP <- univarDeDri$orderframe$IDI[!(rownames(univarDeDriTest$orderframe) %in% colnames(testDataFrame))]
DHCAPTest <- univarDeDriTest$orderframe$IDI[!(rownames(univarDeDriTest$orderframe) %in% colnames(testDataFrame))]
testIDI$DHCAP <- DHCAP
testIDI$DHCAP_T <- DHCAPTest
diffIDI$DHCAP <- DHCAPTest-DHCAP
points(DHCAP,DHCAPTest-DHCAP,pch=4,col=colors[4])
DHCAS <- univarDeDriSpear$orderframe$IDI[!(rownames(univarDeDriSpearTest$orderframe) %in% colnames(testDataFrame))]
DHCASTest <- univarDeDriSpearTest$orderframe$IDI[!(rownames(univarDeDriSpearTest$orderframe) %in% colnames(testDataFrame))]
testIDI$DHCAS <- DHCAS
testIDI$DHCAS_T <- DHCASTest
diffIDI$DHCAS <- DHCASTest-DHCAS
points(DHCAS,DHCASTest-DHCAS,pch=5,col=colors[5])
PCA <- univarPCA$orderframe$IDI[!(rownames(univarPCA$orderframe) %in% colnames(testDataFrame))]
PCATest <- univarPCATest$orderframe$IDI[!(rownames(univarPCA$orderframe) %in% colnames(testDataFrame))]
testIDI$PCA <- PCA
testIDI$PCA_T <- PCATest
diffIDI$PCA <- PCATest-PCA
points(PCA,PCATest-PCA,pch=6,col=colors[6])
EFA <- univarEFA$orderframe$IDI[!(rownames(univarEFA$orderframe) %in% colnames(testDataFrame))]
EFATest <- univarEFATest$orderframe$IDI[!(rownames(univarEFA$orderframe) %in% colnames(testDataFrame))]
testIDI$EFA <- EFA
testIDI$EFA_T <- EFATest
diffIDI$EFA <- EFATest-EFA
points(EFA,EFATest-EFA,pch=7,col=colors[7])
WPCA <- univarWhite$orderframe$IDI[!(rownames(univarWhite$orderframe) %in% colnames(testDataFrame))]
WPCATest <- univarWhiteTest$orderframe$IDI[!(rownames(univarWhite$orderframe) %in% colnames(testDataFrame))]
testIDI$WPCA <- WPCA
testIDI$WPCA_T <- WPCATest
diffIDI$WPCA <- WPCATest-WPCA
points(WPCA,WPCATest-WPCA,pch=8,col=colors[8])
names=c("Raw","HCA:P","HCA:S","DHCA:P","DHCA:S","PCA","EFA","White:PCA")
pchs=c(1,2,3,4,5,6,7,8)
legend("bottomright",names,col=colors,pch=pchs,cex=0.50)

par(op)
Violin of
differences
par(op)
par(mfrow=c(1,1),cex=0.7)
vioplot(diffIDI,
ylim=c(-0.2,0.2),
ylab="Test-Train",
main="Test-Train Paired Predict IDI Difference",
col=colors,
cex.axis=0.6,
las=2
)
stripchart(diffIDI, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(diffIDI),lapply(diffIDI,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
Distribution of
Predict IDI in latent Variables
par(op)
par(mfrow=c(1,1),cex=0.7)
colors2 <- length(testIDI)
colors2[1] <- colors[1];
colors2[2*(1:(length(colors)-1))] <- colors[2:length(colors)]
colors2[1+2*(1:(length(colors)-1))] <- colors[2:length(colors)]
vioplot(testIDI,
ylim=c(0.0,0.5),
ylab="Predict IDI",
main="Predict IDI of Latent Variables",
col=colors2,
cex.axis=0.6,
las=2
)
stripchart(testIDI, method = "jitter", col = "gray",
vertical = TRUE, pch = 1, add = TRUE,cex=0.35)
points(1:length(testIDI),lapply(testIDI,mean), pch = 18, col = "yellow", cex = 2.25)

par(op)
The tables
pander::pander(univarTest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.942 |
6.12688 |
14.011 |
5.53542 |
0.297 |
0.243 |
0.781 |
0.794 |
| RAVLT_immediate |
28.621 |
7.92670 |
37.760 |
10.17622 |
0.459 |
0.144 |
0.689 |
0.765 |
| ADAS11 |
12.972 |
4.60884 |
8.675 |
3.67173 |
0.105 |
0.196 |
0.755 |
0.764 |
| M_ST40CV |
0.179 |
0.00881 |
0.187 |
0.00735 |
0.929 |
0.178 |
0.739 |
0.758 |
| M_ST29SV |
0.125 |
0.00689 |
0.132 |
0.00734 |
0.837 |
0.173 |
0.736 |
0.751 |
pander::pander(univarDeTest$orderframe[1:TopVariables,univariate_columns])
| ADAS11 |
12.97235 |
4.60884 |
8.674712 |
3.67173 |
1.05e-01 |
0.1961 |
0.755 |
0.764 |
| Hippocampus |
0.15807 |
0.00857 |
0.166161 |
0.00928 |
7.63e-01 |
0.1643 |
0.733 |
0.738 |
| FAQ |
5.10606 |
4.92890 |
1.951923 |
2.93201 |
5.70e-12 |
0.1883 |
0.781 |
0.731 |
| La_M_ST24TA |
-0.00142 |
0.00291 |
0.000636 |
0.00278 |
2.98e-01 |
0.0684 |
0.617 |
0.703 |
| M_ST31TA |
0.01867 |
0.00189 |
0.019955 |
0.00166 |
5.36e-01 |
0.1171 |
0.704 |
0.699 |
pander::pander(univarDeSpearTest$orderframe[1:TopVariables,univariate_columns])
| ADAS11 |
12.9723 |
4.60884 |
8.6747 |
3.67173 |
1.05e-01 |
0.1961 |
0.755 |
0.764 |
| M_ST24TA |
0.0259 |
0.00426 |
0.0296 |
0.00378 |
6.49e-02 |
0.1182 |
0.684 |
0.743 |
| FAQ |
5.1061 |
4.92890 |
1.9519 |
2.93201 |
5.70e-12 |
0.1883 |
0.781 |
0.731 |
| M_ST44TA |
0.0206 |
0.00299 |
0.0228 |
0.00313 |
7.31e-01 |
0.0551 |
0.605 |
0.702 |
| M_ST31TA |
0.0187 |
0.00189 |
0.0200 |
0.00166 |
5.36e-01 |
0.1171 |
0.704 |
0.699 |
pander::pander(univarDeDriTest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.9420 |
6.12688 |
14.011 |
5.53542 |
2.97e-01 |
0.243 |
0.781 |
0.794 |
| M_ST29SV |
0.1250 |
0.00689 |
0.132 |
0.00734 |
8.37e-01 |
0.173 |
0.736 |
0.751 |
| FAQ |
5.1061 |
4.92890 |
1.952 |
2.93201 |
5.70e-12 |
0.188 |
0.781 |
0.731 |
| RAVLT_perc_forgetting |
76.3100 |
29.68048 |
52.252 |
32.34899 |
5.27e-02 |
0.104 |
0.685 |
0.712 |
| M_ST52TA |
0.0178 |
0.00174 |
0.019 |
0.00160 |
5.12e-01 |
0.111 |
0.702 |
0.709 |
pander::pander(univarDeDriSpearTest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.9420 |
6.12688 |
14.011 |
5.53542 |
2.97e-01 |
0.243 |
0.781 |
0.794 |
| M_ST29SV |
0.1250 |
0.00689 |
0.132 |
0.00734 |
8.37e-01 |
0.173 |
0.736 |
0.751 |
| FAQ |
5.1061 |
4.92890 |
1.952 |
2.93201 |
5.70e-12 |
0.188 |
0.781 |
0.731 |
| RAVLT_perc_forgetting |
76.3100 |
29.68048 |
52.252 |
32.34899 |
5.27e-02 |
0.104 |
0.685 |
0.712 |
| M_ST52TA |
0.0178 |
0.00174 |
0.019 |
0.00160 |
5.12e-01 |
0.111 |
0.702 |
0.709 |
pander::pander(univarPCATest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.942 |
6.13 |
14.011 |
5.535 |
2.97e-01 |
0.243 |
0.781 |
0.794 |
| RAVLT_immediate |
28.621 |
7.93 |
37.760 |
10.176 |
4.59e-01 |
0.144 |
0.689 |
0.765 |
| ADAS11 |
12.972 |
4.61 |
8.675 |
3.672 |
1.05e-01 |
0.196 |
0.755 |
0.764 |
| FAQ |
5.106 |
4.93 |
1.952 |
2.932 |
5.70e-12 |
0.188 |
0.781 |
0.731 |
| RC2 |
-0.472 |
1.00 |
0.299 |
0.877 |
4.32e-02 |
0.132 |
0.735 |
0.727 |
pander::pander(univarEFATest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.942 |
6.127 |
14.011 |
5.535 |
2.97e-01 |
0.243 |
0.781 |
0.794 |
| RAVLT_immediate |
28.621 |
7.927 |
37.760 |
10.176 |
4.59e-01 |
0.144 |
0.689 |
0.765 |
| ADAS11 |
12.972 |
4.609 |
8.675 |
3.672 |
1.05e-01 |
0.196 |
0.755 |
0.764 |
| MR2 |
-0.465 |
0.969 |
0.295 |
0.863 |
1.51e-02 |
0.127 |
0.726 |
0.732 |
| FAQ |
5.106 |
4.929 |
1.952 |
2.932 |
5.70e-12 |
0.188 |
0.781 |
0.731 |
pander::pander(univarWhiteTest$orderframe[1:TopVariables,univariate_columns])
| ADAS13 |
20.94 |
6.127 |
14.01 |
5.535 |
2.97e-01 |
0.243 |
0.781 |
0.794 |
| RAVLT_immediate |
28.62 |
7.927 |
37.76 |
10.176 |
4.59e-01 |
0.144 |
0.689 |
0.765 |
| ADAS11 |
12.97 |
4.609 |
8.67 |
3.672 |
1.05e-01 |
0.196 |
0.755 |
0.764 |
| FAQ |
5.11 |
4.929 |
1.95 |
2.932 |
5.70e-12 |
0.188 |
0.781 |
0.731 |
| L2 |
39.78 |
0.977 |
40.55 |
0.898 |
3.36e-01 |
0.123 |
0.711 |
0.729 |
topUni <- univar$orderframe$Name[1:TopVariables]
topDe <- univarDe$orderframe$Name[1:TopVariables]
topDeSpear <- univarDeSpear$orderframe$Name[1:TopVariables]
topDeDri <- univarDeDri$orderframe$Name[1:TopVariables]
topDeDriSpear <- univarDeDriSpear$orderframe$Name[1:TopVariables]
topPCA <- univarPCA$orderframe$Name[1:TopVariables]
topEFA <- univarEFA$orderframe$Name[1:TopVariables]
topPCAWhite <- univarWhite$orderframe$Name[1:TopVariables]
Model of top
variables
par(mfrow=c(1,2),cex=0.6)
lmRAW <- glm(paste(outcome,"~."),
trainDataFrame[,c(outcome,topUni)],
family="binomial")
prRaw <- predictionStats_binary(cbind(testDataFrame[,outcome],predict(lmRAW,testDataFrame)),"Top Raw",cex=0.75)
Top Raw
lmDe <- glm(paste(outcome,"~."),
DEdataframe[,c(outcome,topDe)],
family="binomial")
prDe <- predictionStats_binary(cbind(predTestDe[,outcome],predict(lmDe,predTestDe)),"Top HCA:P",cex=0.75)
Top HCA:P

lmDeSpear <- glm(paste(outcome,"~."),
DEdataframeSpear[,c(outcome,topDeSpear)],
family="binomial")
prSpear <- predictionStats_binary(cbind(predTestDeSpear[,outcome],predict(lmDeSpear,predTestDeSpear)),"Top HCA:S",cex=0.75)
Top HCA:S
lmDri <- glm(paste(outcome,"~."),
DriDEdataframe[,c(outcome,topDeDri)],
family="binomial")
prDri <- predictionStats_binary(cbind(predTestDe[,outcome],predict(lmDri,predTestDri)),"Top DHCA:P",cex=0.75)
Top DHCA:P

lmDriSpear <- glm(paste(outcome,"~."),
DriDEdataframeSpear[,c(outcome,topDeDriSpear)],
family="binomial")
prDriSpear <- predictionStats_binary(cbind(predTestDriSpear[,outcome],predict(lmDriSpear,predTestDriSpear)),"Top DHCA:S",cex=0.7)
Top DHCA:S
lmPCA <- glm(paste(outcome,"~."),
PCA_Train[,c(outcome,topPCA)],
family="binomial")
prPCA <- predictionStats_binary(cbind(PCA_Predicted[,outcome],predict(lmPCA,PCA_Predicted)),"Top PCA",cex=0.75)
Top PCA

lmEFA <- glm(paste(outcome,"~."),
EFA_Train[,c(outcome,topEFA)],
family="binomial")
prEFA <- predictionStats_binary(cbind(EFA_Predicted[,outcome],predict(lmEFA,EFA_Predicted)),"Top EFA",cex=0.75)
Top EFA
lmPCAW <- glm(paste(outcome,"~."),
PCAWhite_Train[,c(outcome,topPCAWhite)],
family="binomial")
prWPCA <- predictionStats_binary(cbind(PCAWhitePredicted[,outcome],predict(lmPCAW,PCAWhitePredicted)),"Top White:PCA",cex=0.75)
Top White:PCA

par(op)
The Performance
Tables and Plots
par(cex=0.6)
aucs <- prRaw$aucs
aucs <- rbind(aucs,prDe$aucs)
aucs <- rbind(aucs,prSpear$aucs)
aucs <- rbind(aucs,prDri$aucs)
aucs <- rbind(aucs,prDriSpear$aucs)
aucs <- rbind(aucs,prPCA$aucs)
aucs <- rbind(aucs,prEFA$aucs)
aucs <- rbind(aucs,prWPCA$aucs)
rownames(aucs) <- c("RAW",
"HCA:P",
"HCA:S",
"DHCA:P",
"DHCA:S",
"PCA",
"EFA",
"WPCA"
)
pander::pander(aucs)
| RAW |
0.816 |
0.769 |
0.863 |
| HCA:P |
0.841 |
0.798 |
0.885 |
| HCA:S |
0.828 |
0.782 |
0.874 |
| DHCA:P |
0.844 |
0.800 |
0.887 |
| DHCA:S |
0.844 |
0.800 |
0.887 |
| PCA |
0.840 |
0.794 |
0.885 |
| EFA |
0.837 |
0.792 |
0.883 |
| WPCA |
0.843 |
0.799 |
0.888 |
bpAUC <- barPlotCiError(as.matrix(aucs),
metricname = "ROC AUC",
thesets = "Test AUC",
themethod = rownames(aucs),
main = "ROC AUC",
offsets = c(0.5,1),
scoreDirection = ">",
ho=0.5,
args.legend = list(bg = "white",x="bottomleft",inset=c(0.0,0),cex=0.5),
col = terrain.colors(nrow(aucs))
)

berror <- prRaw$berror
berror <- rbind(berror,prDe$berror)
berror <- rbind(berror,prSpear$berror)
berror <- rbind(berror,prDri$berror)
berror <- rbind(berror,prDriSpear$berror)
berror <- rbind(berror,prPCA$berror)
berror <- rbind(berror,prEFA$berror)
berror <- rbind(berror,prWPCA$berror)
rownames(berror) <- rownames(aucs)
pander::pander(berror)
| RAW |
0.269 |
0.217 |
0.322 |
| HCA:P |
0.235 |
0.187 |
0.285 |
| HCA:S |
0.257 |
0.208 |
0.309 |
| DHCA:P |
0.241 |
0.187 |
0.291 |
| DHCA:S |
0.241 |
0.189 |
0.291 |
| PCA |
0.220 |
0.173 |
0.268 |
| EFA |
0.239 |
0.195 |
0.287 |
| WPCA |
0.241 |
0.193 |
0.291 |
bpBER <- barPlotCiError(as.matrix(berror),
metricname = "Balanced Error Rate",
thesets = "Test BER",
themethod = rownames(aucs),
main = "Balanced Error Rate",
offsets = c(0.5,1),
scoreDirection = "<",
ho=0.5,
args.legend = list(bg = "white",x="topleft",inset=c(0.0,0),cex=0.5),
col = terrain.colors(nrow(aucs))
)

par(op)